Market Risk

FRM Part II - Book 1

A coherent risk measure is a risk metric that satisfies a set of properties, such as subadditivity, positive homogeneity, and translation invariance.

What is Subadditivity?

• Subadditivity: The risk measure of a portfolio is less than or equal to the sum of the risk measures of its components.

A coherent risk measure is a risk metric that satisfies a set of properties, such as subadditivity, positive homogeneity, and translation invariance.

What is Positive homogeneity ?

• Positive homogeneity: Scaling the portfolio by a positive factor scales the risk measure by the same factor.

A coherent risk measure is a risk metric that satisfies a set of properties, such as subadditivity, positive homogeneity, and translation invariance.

What is translation invariance?

• Translation invariance: Adding a constant to the portfolio does not change the risk measure.

A generally coherent risk measure tends to involve increasingly sophisticated weighting functions. Which of the following is a suitable replacement for the equal weights in the 'average VaR' to estimate any risk measure?

Weights appropriate to risk measure being estimated.

The 'average VaR' is a simple method for estimating the VaR of a portfolio or assets by averaging the VaR of each component, with equal weights assigned to each component. However, because different components may have varying levels of risk, correlation, or impact on the overall portfolio, this approach may not be appropriate for all types of risk measures. Therefore, the appropriate weights must be chosen based on the underlying component characteristics and the specific risk measure being evaluated to estimate any risk measure using a weighted approach

Coherent risk measures are fundamental in the field of risk management, providing a consistent framework for assessing the risk of financial portfolios. The key to estimating coherent risk measures lies in the:

Ability to estimate quantiles.

Many coherent risk measures, particularly Expected Shortfall (ES), depend on accurately estimating the tail of the loss distribution. Expected Shortfall is defined as the average loss beyond a specified quantile (VaR). Therefore, before ES can be calculated, the corresponding quantile (VaR) must first be estimated.

An analyst has gathered the following information about a portfolio that has normally distributed geometric returns:

Mean: 10%
Standard Deviation: 40%
Portfolio: 100 million

What is the 95% lognormal VAR for this portfolio?

$42.8 million.

Assume you are dealing with a stock “ A” that displays a highly negatively skewed distribution comprised of the past 260-days returns. Suppose you have P1 = A and P2 = -A, meaning P1 is long stock A and P2 is short stock A. Which statement is most likely to be accurate about a 99% VaR?

Correct Answer is #1 - |VaR (P1)| > |VaR (P2)|.

Given that the return distribution of stock A is negatively skewed, it displays a long left tail. This implies large potential losses for a long position and large potential gains for a short position. Therefore, |VaR (P1)| will be expected to be higher.

A data scientist is analyzing a dataset and wants to determine the distribution of his data. The scientist decides to use a QQ plot in his analysis. Which of the following statements about QQ plots is correct?

Option 3 - QQ plots are useful in determining if a dataset follows a normal distribution.

A QQ plot is a graphical tool used in statistics to help visualize how a dataset is distributed. It plots the quantiles of the dataset against the quantiles of a standard normal distribution. If the dataset follows a normal distribution, the points on the QQ plot will approximately lie on a straight line. This is because the quantiles of the dataset will match closely with the quantiles of the standard normal distribution. Therefore, a QQ plot is a powerful tool for visually checking the assumption of normality in a dataset.

A small prop trading firm has only 80 days of historical returns. They want to estimate VaR at 97.5% but find the traditional historical simulation percentile approach too coarse. The head quant suggests a non-parametric density estimation method.

What key benefit does this non-parametric approach offer the firm?

Option 2 - Flexible percentile selection.

The key benefit of using a non-parametric density estimation method like Kernel Density Estimation (KDE) is that it creates a smooth, continuous distribution from the limited historical data. This allows for flexible percentile selection.

With only 80 data points, using a direct percentile method (like taking the 2nd smallest return for a 97.5% VaR) is very sensitive to individual data points and doesn't give a very precise estimate. KDE interpolates between the data points, creating a much smoother distribution from which you can read off any percentile you want with greater precision than the raw data would allow.

Which statement correctly captures the essence of a correlation-weighted approach within the historical simulation framework?

Option 2 - Correlation-weighted historical simulation aims to address a key limitation of standard historical simulation: its reliance on historical correlations, which may not be representative of current or future market conditions. This approach adjusts the historical returns in a way that preserves the individual volatilities of the assets but alters their co-movements (correlations) to reflect a desired or estimated correlation structure. This is often done using techniques like Cholesky decomposition.

Even though bootstrapping has numerous advantages, the bootstrap estimates are associated with a little bias or error. Which of the following presents an error of bootstrapping?

Re-sampling variability

In the context of bootstrapping, re-sampling variability refers to the error that arises from the fact that we take only a finite number of bootstrap re-samples (denoted by B) from our original sample, rather than an infinite number. This limitation is inherent to the bootstrapping process, as it is practically impossible to take an infinite number of re-samples. As a result, our bootstrap estimates are subject to a certain degree of variability that is directly related to the number of re-samples taken. The larger the number of re-samples, the lower the re-sampling variability, and vice versa. However, even with a large number of re-samples, the re-sampling variability can never be completely eliminated, thus contributing to the overall error in the bootstrap estimates.

Paul is using the age-weighted historical simulation approach to estimate the VaR of a stock portfolio, with a historical sample size of 100 days and a decay factor of 0.96. Over the recent past, the portfolio has registered the following returns.

Determine the weight on the return earned 45 days ago

0.006751.

An analyst performing a historical simulation to measure the VaR of a portfolio uses a volatility-weighted approach. One month ago, the actual return of the asset was 5% and its daily volatility estimate was 2%. If the current daily volatility is 1.5%, calculate the volatility-adjusted return.

0.0375

Understanding the correlation between portfolio assets is crucial in risk management, especially when calculating Value at Risk (VaR). The degree of correlation significantly impacts the overall risk profile of the portfolio. If portfolio assets are perfectly correlated, portfolio VaR will equal:

Undiversified VaR

When the assets in a portfolio are perfectly correlated, the portfolio's VaR would be equal to the undiversified VaR. This is because perfect correlation implies that all assets in the portfolio move in the same direction at the same time. Therefore, there is no benefit from diversification, and the portfolio's risk is equivalent to the sum of the individual asset risks, which is the definition of undiversified VaR.

A large investment bank's risk management team is implementing a VaR mapping framework for their diversified fixed-income portfolio containing corporate bonds, government securities, and mortgage-backed securities across multiple currencies and credit ratings. The portfolio exhibits significant exposure to both systematic market movements and idiosyncratic issuer-specific events. The risk manager must decide on the appropriate mapping methodology that will most effectively capture the dual nature of risks while maintaining computational efficiency for daily risk reporting. The team recognizes that the chosen approach will fundamentally determine how well their VaR estimates reflect the portfolio's true risk profile during both normal and stressed market conditions.

Which approach would most appropriately balance the capture of general and specific risks in this mapping process?

Map each security to principal component factors plus issuer-specific idiosyncratic risk components

The principal component approach with issuer-specific idiosyncratic components most effectively captures both general and specific risks in VaR mapping. Principal components capture the systematic risk factors that drive correlated movements across the fixed-income universe (general risk), while the issuer-specific idiosyncratic components capture the unique risks associated with individual securities (specific risk). This approach provides the optimal balance between comprehensiveness and computational efficiency, as principal components reduce dimensionality while preserving most of the variance in the data, and the idiosyncratic components ensure that security-specific events are not overlooked in the risk calculation.

Adding more general risk factors to a VaR model will most likely:

2 - Decrease the size of specific risks.

The addition of more general risk factors to a Value at Risk (VaR) model will most likely decrease the size of specific risks. Specific risk, also known as idiosyncratic risk or residual risk, refers to the risk associated with individual assets in a portfolio, which is not correlated with market risk. By incorporating more general risk factors into the model, the model becomes more comprehensive and can better capture the overall risk dynamics of the portfolio. This, in turn, reduces the portion of risk that is unexplained by the model, i.e., the specific risk. For instance, if a model only considers the duration of the portfolio, it will have a larger specific risk compared to a model that considers both duration and credit risk. Therefore, adding more general risk factors to the model helps define the risk more accurately and reduces the specific or residual risk.

A forward rate agreement is a type of forward contracts that allows the contracting parties to make transactions in a locked interest rate at some future date. The buyer in this contract locks in the borrowing rate and the seller locks in the lending rate at some future date.

Which statement is true about such a contract if the spot rate is higher than the forward rate at the time of the transaction?

Option 3 - The buyer will benefit by borrowing at a lower locked-in rate, while the seller will be worse off by lending at a lower rate.

In a Forward Rate Agreement (FRA), the buyer (or the 'long') locks in a borrowing rate, while the seller (or the 'short') locks in a lending rate. If the spot rate at the time of the transaction is higher than the forward rate, the buyer will benefit. This is because the buyer will make payments at a lower rate than the prevailing market rate. On the other hand, the seller will be worse off because they will be lending at a lower rate than the current market rate. This scenario is advantageous for the buyer because they have effectively hedged against the risk of rising interest rates. Conversely, the seller, who may have anticipated a fall in interest rates, will be at a disadvantage as they are locked into lending at a lower rate.

Benchmarking is the process of evaluating a portfolio’s risk against some standard or ideal portfolio risk that is considered as the benchmark. After we performed the necessary calculations for portfolio A, we found the tracking error VaR of portfolio A which is 0.63 million. What does this tracking error VaR value imply?

Option 1 - The maximum deviation between the index and portfolio A is at most 0.63 million under normal market conditions.

The Tracking Error Value at Risk (TE-VaR) is a measure of the maximum potential deviation between the portfolio and the benchmark index under normal market conditions. In this case, the TE-VaR of 0.63 million implies that the maximum deviation between the index and portfolio A is at most 0.63 million under normal market conditions. This is a measure of the risk of underperformance relative to the benchmark. It is important to note that 'normal market conditions' refer to the usual, everyday fluctuations in the market, not extreme events or crises. The TE-VaR is a critical tool for portfolio managers as it helps them understand the potential risk of their portfolio underperforming the benchmark, allowing them to make informed decisions about risk management and asset allocation.

Considering an example of a two-bond portfolio, we calculated the portfolio returns and the risks associated with those portfolios using the mapping technique. Then, we found some specific values, say, 2.80 VaR for duration mapping and 2.67 VaR for cash flow mapping. This notable difference in these values is due to the fact that:

Option 1 - risk measures are not perfectly linear with maturity and correlations are below unity.

The difference in VaR values for duration mapping and cash flow mapping is due to two factors. The first factor is that risk measures are not perfectly linear with maturity. This means that as the maturity of the bond increases, the risk does not increase at a constant rate. Instead, it may increase at a decreasing rate, remain constant, or even decrease.

The second factor is that correlations are below unity. A correlation below unity (less than 1) means that the variables do not move perfectly in sync with each other. This reduces the overall risk of the portfolio because the negative performance of one bond may be offset by the positive performance of the other bond. Therefore, both these factors contribute to the difference in VaR values for duration mapping and cash flow mapping.

As we know, mapping is the simple process of analyzing the market instruments on primitive risk factors. Considering this concept, let’s take a single instrument that has a market value of Vm. It is allocated to specific risk exposures namely X1, X2, and X3. Suppose not all of the current market value Vm is allocated to these risk factors, and there are no omitted factors. What does that imply with regard to the remaining value?

Option 3 - The remainder value is allocated to cash which is not a risk factor.

In this scenario, if the entire market value Vm of the instrument is not allocated to the risk factors X1, X2, and X3, it implies that the remaining value is held in cash. This portion of the value is not exposed to any risk factors because cash is not subject to market risk, credit risk, or other types of financial risk. Therefore, it is accurate to say that the remaining value is allocated to cash, which is not a risk factor.

The peaks-over-threshold approach is a method used in extreme value theory to analyze the tail behavior of a distribution. This technique is particularly useful for assessing the risk of rare and extreme events. The peaks-over-threshold approach generally requires:

Option 1 - fewer estimated parameters than the GEV approach and shares one parameter with the GEV.

The GEV distribution models block maxima and requires three parameters: the location parameter, the scale parameter, and the shape parameter, which describes the tail behavior. In contrast, the POT approach focuses only on exceedances over a high threshold, requiring two parameters: the scale parameter and the same shape parameter. Note that only one parameter — the shape parameter ξ — is truly shared between the two approaches in the sense of being the same quantity with the same interpretation.

In practice, risk analysts prefer the Peaks-over-threshold (POT) approach over the generalized extreme value approach because the POT approach - why is this?

Is more efficient in the use of data

The efficiency of the POT approach stems from its ability to focus on the distribution of exceedances over a threshold, which is particularly useful for the risk measurement of tail losses. In contrast, the GEV approach involves some loss of useful data relative to the POT approach, because some blocks might have more than one extreme in them. Therefore, the POT approach's efficient use of data makes it a preferred choice among risk analysts.

The Fréchet distribution is a type of extreme value distribution used to model the behavior of extreme values, such as maximum or minimum values in a dataset. The shape parameter ξ in the Fréchet distribution determines the tail behavior of the distribution.

What does a higher shape distribution mean?

A higher ξ corresponds to heavier tails.

The Generalized Extreme Value (GEV) distribution is a family of continuous probability distributions developed within extreme value theory. It combines the Gumbel, Fréchet and Weibull families, also known as type I, II and III extreme value distributions. By varying the parameter ξ, the GEV distribution can model these three types of distributions.

What is the Weibull distrubution?

When ξ < 0, the GEV becomes the Weibull distribution. The Weibull distribution has a cumulative distribution function with lighter than normal tails, meaning that extreme outcomes are less likely than under a normal distribution.

What is the Frechet distribution?

The Generalized Extreme Value (GEV) distribution becomes the Frechet distribution when the shape parameter ξ is greater than zero. The Frechet distribution is a particular case of the GEV distribution and is used to model the distribution of extreme events. It is heavy-tailed and is appropriate for modelling the maximum values of distributions with a heavy right tail.

This means it is often used in fields such as hydrology, telecommunications, finance, and insurance to model extreme events such as large floods, large insurance claims, or large stock market movements.

An analyst is reviewing a bank’s backtesting framework. For a 1-day VaR model, the analyst wants to isolate model performance from the effects of trading activity, fees, and interest accruals. Which two return series should the analyst use?

Option 1 - The actual returns and the hypothetical returns.

Actual returns measure the real P&L of the trading portfolio, including all trading activity, fees, commissions, and net interest. Hypothetical returns are based on the portfolio frozen at the start of the day, excluding any intraday trading or operational changes. Comparing both datasets allows the analyst to identify whether exceptions are due to trading/operational effects or the model itself.

Using a two‑tailed 95% backtest over 250 days, compare the nonrejection band (the set of exception counts that do not trigger rejection) for 95% vs. 99% VaR. Which statement is correct?

Option 1 - The nonrejection band in exception counts is narrower for 99% VaR than for 95% VaR.

p> Exception counts are discrete. With a small tail probability (99% VaR), the expected count is near 2–3, so the feasible acceptance region spans few integers. At 95% VaR, the expected count is around 12–13, yielding a wider integer range for nonrejection. Hence, the 99% VaR band is narrower.

A bank currently estimates its 1-day VaR at the 99% confidence level. It is considering lowering this to 95% to improve the effectiveness of its backtesting process. Which of the following statements is correct?

Option 1 - The backtest will have higher power at 95% VaR than at 99% VaR.

Lowering the VaR confidence from 99% to 95% increases the expected exception rate (from 1% to 5%). With more tail observations in a fixed sample, backtesting gains statistical power to detect model misspecification. The test becomes more informative because exception counts are less sparse.

When a backtest is conducted with a short sample period, the most significant statistical limitation is:

Option 1 - Low test power due to high variability in exception counts.

Short samples result in large sampling error around the expected exception rate, making it difficult to distinguish between a correct model and an incorrect one. Larger samples provide more reliable backtesting results.

A bank’s 95% VaR backtest over 252 days shows 13 exceptions overall, but 10 occurred within a two-week sell-off. Risk control argues the annual count is on target; the model validation team worries about timing. Which action best addresses the deficiency implied by this pattern?

Option 1 - Apply Christoffersen’s conditional coverage (LR_cc) to test both correct frequency and independence of exceptions.

Christoffersen’s framework augments LR_uc with an independence test (LR_ind) and combines them (LR_cc) to detect time-varying risk/“bunching.” This directly targets the dependence structure that simple counts (i.e frequency only tests) miss.

Emma Chen, a risk manager at an investment firm, is backtesting a daily 99% Value-at-Risk (VaR) model over a 5-year period at a 95% confidence level. Assuming 252 trading days per year and that daily returns are independently and identically distributed, which of the following is closest to the maximum number of daily losses exceeding the 1-day 99% VaR in 5 years that is acceptable to conclude that the model is calibrated correctly?

• Expected Exceptions:

  • Probability of an exception for a 99% VaR: p=0.01.

  • Number of days: n=5×252=1260.

  • Expected exceptions: μ=n⋅p=1260⋅0.01=12.6.

• Variance and Standard Deviation:

  • Variance: σ2=n⋅p⋅(1−p)=1260⋅0.01⋅0.99=12.474.

  • Standard deviation: σ=12.474√≈3.5319.

• 95% Confidence Interval (Normal Approximation):

  • For 95% confidence, use z=1.96.

  • Confidence interval: μ±z⋅σ=12.6±1.96⋅3.5319≈12.6±6.9225.

  • Lower bound: 12.6−6.9225≈5.6775.

  • Upper bound: 12.6+6.9225≈19.5225.

  • Integer acceptance region: 6≤X≤19 (since 5.6775≈6 and 19.5225≈19).

• Maximum Acceptable Number of Exceptions:

  • The maximum number of daily losses (exceptions) that allows the model to be considered calibrated is the upper bound of the acceptance region, which is 19.

While validating a bank’s VaR model, an analyst observes that the model passes with actual returns but fails with hypothetical returns. Which conclusion is most appropriate?

Option 3 - The model is inadequate and likely has a methodological problem.

Hypothetical returns isolate the VaR model’s performance by excluding intraday trades and operational factors. If the model fails with hypothetical returns, the issue lies in the model’s assumptions, data inputs, or methodology rather than in trading activities. This points to the need for recalibration or redesign of the model.

In regulatory backtesting for a 99% VaR model over 250 days, the green zone corresponds to an observed number of exceptions of:

0 to 4 exceptions - For a 99% VaR model over 250 days, the green zone covers 0–4 exceptions, indicating no penalty multiplier increase is applied to capital requirements.

Yellow (5-9) and red zones (10 or more) increase multipliers as exceptions rise.

A risk analyst is reviewing the backtesting results of a VaR model and interpreting the statistical tests used to evaluate model accuracy. The analyst must distinguish between potential errors in the backtesting process to accurately assess the model’s performance. Which statement correctly identifies a Type II error in the context of VaR backtesting?

Accepting an incorrect VaR model as accurate.

Type II error (False negative): Failing to reject the model when it is actually inaccurate. This can lead to underestimating risk.

What is a type I error?

Type I error (False positive): Rejecting a correct VaR model (i.e., concluding the model is inaccurate when it is actually accurate)

Matthew Hopkins is invited to interview for a position as a financial risk manager. After completing an initial set of questions, the interviewer asks for the interpretation of the following case: a $20 million 15-day VAR figure having a confidence level of 95%. Which of the following represents the CORRECT interpretation?

Option 3 - The amount of loss spread over the next 15 days is expected to be less than $20 million in 95 percent of case scenarios.

• Value at Risk (VaR) is a measure of the potential loss in value of a portfolio or investment over a specific time period and with a certain level of confidence.

• A 95% confidence level in VaR means that there is a 5% chance that the actual loss will exceed the VaR estimate.

• It is important to consider the assumptions and limitations of VaR when using it for risk management and decision-making i.e VaR models don’t capture changing portfolios and the sample backtested may not be representative of the true underlying risk.

The VaR Quantile Regression (VQR) test evaluates whether the VaR model correctly predicts quantiles of the loss distribution across a range of market conditions. It is a more granular and demanding test compared to other backtesting methods.

What does a significantly higher failure rate observed for the VQR test mean?

The significantly higher failure rate observed with the VQR test suggests that it may be identifying subtle deviations in model performance or reacting to non-stationary data in the portfolio (e.g., changes in risk factors or regime shifts). This does not necessarily imply that the model is fundamentally flawed but highlights its sensitivity.

A regulator observes that a group of 20 banks, subject to market risk rules, have an average exceedance rate of 0.4% for their 99% VaR models during a period from 2013-2016. Given this period is described as "benign for markets," which of the following is the MOST likely interpretation of this observation in the context of model validation?

Option 2 - The low exceedance rate, while seemingly positive, may indicate excessive conservatism in the models, potentially leading to inefficient capital allocation.

Prior to 2013, some banks included fees, commissions, and intraday trading revenue in their profit and loss (P&L) calculations used for VaR backtesting. What potential problem did this practice introduce?

Option 1 - It could have masked actual portfolio losses, making the VaR model appear more accurate than it was.

Including fees, commissions, and intraday trading revenue in P&L calculations could distort the true relationship between actual losses and the VaR estimate. These items are unrelated to portfolio risk but can offset trading losses, resulting in fewer observed exceptions during backtesting. This practice would make the VaR model seem more accurate than it truly was by underrepresenting the frequency and magnitude of exceptions.

Backtesting Value-at-Risk (VaR) models can face challenges due to limited data, especially when considering rare events. This can lead to low statistical power, making it difficult to reject inaccurate models. Which of the following strategies is MOST effective in addressing this power issue in backtesting?

Option 3 - Incorporating additional information beyond simple exception counts as is done in Dynamic Quantile (DQ) and Logistic Dynamic Quantile (LDQ) tests.

Traditional backtesting methods like the unconditional coverage test rely solely on exception counts, which can lead to low statistical power, especially with small datasets. Advanced methods such as the DQ and LDQ tests incorporate additional information, including the timing, clustering, and dependence of exceptions. By leveraging more nuanced data, these tests enhance the ability to identify poorly specified models, addressing the statistical power issue effectively.

Note while a rolling window approach increases the number of backtesting observations, it introduces autocorrelation in the test results, violating the independence assumption of many backtesting tests and potentially leading to spurious conclusions. It doesn't directly address the power of the tests themselves.

In backtesting VaR models, a critical consideration is the balance between Type I and Type II errors. In the context of risk management:

Option 2 - Type II error is generally more costly

Type II error: Failing to reject a false null hypothesis (failing to reject an incorrect model). In risk management, a Type II error—failing to identify a model that underestimates risk—is generally considered more dangerous. This could lead to insufficient capital reserves and potentially catastrophic losses if the true risk is higher than the model suggests. A Type I error, while inconvenient, primarily results in unnecessary model adjustments or further investigation of a sound model.

A risk analyst is comparing different methods for constructing confidence intervals for one-day 99% VaR estimates of the S&P 500. They are considering historical simulation, GARCH, and filtered historical simulation (FHS) VaR, along with various confidence interval estimation techniques, including order statistics and bootstrapping. Based on empirical studies, which of the following statements regarding the width of confidence intervals is correct?

Option 2- Filtered historical simulation (FHS) tends to produce narrower confidence intervals compared to historical simulation, particularly during stress periods.

Filtered historical simulation adjusts historical returns by accounting for changing volatility (typically modelled using GARCH or similar methods). This reduces the noise in historical data, particularly during periods of market stress, leading to more stable and narrower confidence intervals compared to standard historical simulation, which relies purely on raw historical data.

A risk manager is using historical simulation to calculate VaR and is considering using bootstrap techniques to create confidence intervals. The historical dataset is relatively small. Which of the following is a key limitation of bootstrapping in this scenario?

Option 1 - The bootstrapped samples may not adequately capture extreme tail events.

Bootstrapping relies on resampling from the available historical dataset to estimate the distribution of possible outcomes. When the dataset is small, the likelihood of capturing rare, extreme tail events diminishes because these events occur infrequently in the historical data. This limitation affects the reliability of the confidence intervals for VaR, especially for portfolios sensitive to tail risk.

A financial institution, Alpha Investments, is comparing two VaR models, A and B, using the signs test. Which of the following correctly describes the key focus of the sign test in this context?

Option 3 - Determining if the median loss differential between the two models is significantly different from zero.

The sign test is a non-parametric statistical test used to evaluate whether the median of the differences between two paired samples (in this case, the losses predicted by the two VaR models) is significantly different from zero. In the context of comparing VaR models, it checks whether one model consistently produces higher or lower losses compared to the other, without assuming a specific distribution for the differences.